Académique Documents
Professionnel Documents
Culture Documents
Thesis
by Martina Jetzki
Department of Airport and Air Transportation Research
RWTH AACHEN UNIVERSITY
23.12.2009
Written at:
Contact:
EUROCONTROL
martina.jetzki@rwth-aachen.de
Rue de la Fussee 96
1140 Brussels, Belgium
Supervisor from EUROCONTROL:
Philippe Enaud, Deputy Head of Unit (PRU)
Yves De Wandeler, FTA-CODA
Supervisor from RWTH Aachen:
Univ.-Prof. Dr. rer. nat. Johannes Reichmuth
Dipl.-Wi.-Ing. Sebastian Kellner
Acknowledgement
Firstly, I would like to thank Prof. Dr. Reichmuth who gave me the opportunity
and Sebastian Kellner who encouraged me in the first place, to write this thesis at
EUROCONTROL, Brussels.
I am very grateful for the amazing assistance and lasting mentoring I experienced
from EUROCONTROL staff. I thank Dr. David Marsh and Philippe Enaud for
counselling me with ideas and advice. In addition, Id like to express my
gratefulness to Yves De Wandeler who is not only a genuine expert in delay
analysis, but who also kindly assisted me with helpful advice in all matters during
this whole period; Magda Gregorova my personal SAS assistant and Holger
Hegendrfer for his encouragement and support especially in stressful times.
It was a real pleasure working in this multicultural, multilingual and above all
inspiring environment.
Abstract
This empirical study is concerned with the propagation of delays in European air
traffic. The so called reactionary delays account for about 40 percent of all
departure delays in Europe but, due to data limitations, most delay studies have
traditionally focused on the analysis of primary delays at the departure airports.
Using data collected by the Central Office for Delay Analysis (CODA), this study
developed aircraft sequences in order to analyse the propagation of delays and
to better understand the amplifying or mitigating factors.
Hub-and-spoke carriers tend to have a smaller level of propagation than point-topoint and low-cost carriers because they have a higher ability to absorb delay
during the ground phases. On the other hand, low-cost operations absorb notably
more delay in the block phase than the other operations.
Overall, the sequences of reactionary delays starting in the morning have a
higher impact and magnitude than the ones starting in the afternoon as they
propagate on average on more subsequent flight legs.
However, the level of propagation in the afternoon appears to be higher which
suggests that airline efforts to mitigate delay propagation are higher in the
morning than in the afternoon. Moreover, the magnitude of sequences of
reactionary delays after short delays is higher, because reactionary delays
increase throughout the sequence due to further primary delays in block and
ground phase.
Looking at major European hubs, it was observed that they affect daily 30 to 50
other airports, but in terms of reactionary delays they mostly affect their own
operations. Aircraft returning to the hub after one flight leg arrive with up to 50
percent of the original departure delay when leaving the hub airport.
TABLE OF CONTENTS
1
2
3
6
7
INTRODUCTION..................................................................................................................1
1.1
BACKGROUND ................................................................................................................1
1.2
OBJECTIVE ....................................................................................................................5
1.3
STUDY SCOPE ................................................................................................................6
1.3.1 GEOGRAPHICAL SCOPE ...................................................................................................6
1.3.2 TEMPORAL SCOPE ..........................................................................................................6
1.4
ORGANISATION OF THE STUDY .........................................................................................7
LITERATURE REVIEW .......................................................................................................8
DATA VALIDATION & PROCESSING ..............................................................................13
3.1
DATA SOURCES ...........................................................................................................13
3.1.1 CENTRAL FLOW MANAGEMENT UNIT (CFMU).................................................................13
3.1.2 CENTRAL ROUTE CHARGES OFFICE (CRCO) .................................................................13
3.1.3 CENTRAL OFFICE FOR DELAY ANALYSIS (CODA)............................................................13
3.2
DATA VALIDATION & LIMITATIONS ..................................................................................20
3.2.1 MISSING OR INCOMPLETE DATA ......................................................................................20
3.2.2 USE OF DIFFERENT DELAY CODES ..................................................................................21
3.2.3 DIFFERENT CODING POLICIES.........................................................................................22
3.2.4 ERRORS IN DATASETS ...................................................................................................23
3.2.5 MISSING FLIGHTS ..........................................................................................................24
3.3
INPUT IN ANALYSIS........................................................................................................24
3.4
DATA PROCESSING ......................................................................................................25
3.4.1 BUILDING SEQUENCES WITH AIRLINE ROTATIONS .............................................................25
3.4.2 GROUPING BY AIRLINE BUSINESS MODEL.........................................................................27
3.4.3 CONVERTING UNIVERSAL TIME COORDINATED (UTC) .....................................................29
CONCEPTUAL FRAMEWORK .........................................................................................30
4.1
FACTORS DETERMINING THE LEVEL OF REACTIONARY DELAY ...........................................30
4.2
KPIS OF REACTIONARY DELAYS.....................................................................................30
4.2.1 SENSITIVITY TO PRIMARY DELAYS IN AIRLINE BUSINESS MODELS .......................................31
4.2.2 AIRLINE SCHEDULING MATTERS ......................................................................................31
4.3
SEQUENCE OF FLIGHTS WITH REACTIONARY DELAYS .......................................................35
4.3.1 CREATING SEQUENCES OF SUBSEQUENT FLIGHT LEGS WITH REACTIONARY DELAYS ............35
4.3.2 ROOT DELAY ................................................................................................................36
4.3.3 DEPTH OF THE SEQUENCE .............................................................................................36
4.3.4 MAGNITUDE .................................................................................................................36
ANALYSIS OF REACTIONARY DELAYS.........................................................................38
5.1
DISTRIBUTION OF PRIMARY DELAYS BY DURATION ...........................................................39
5.2
SENSITIVITY OF AIRLINE BUSINESS MODELS TO REACTIONARY DELAY ...............................39
5.2.1 METHODS OF CALCULATING REACTIONARY DELAY ...........................................................39
5.2.2 SHARE OF REACTIONARY DELAY BY TYPE OF OPERATION ..................................................41
5.3
ABILITY TO ABSORB REACTIONARY DELAYS IN THE BLOCK-TO-BLOCK PHASE ....................49
5.4
ABILITY TO ABSORB REACTIONARY DELAYS IN THE TURN-AROUND PHASE .........................54
5.4.1 DELAY DIFFERENCE INDICATOR-GROUND AND GROUND TIME OVERSHOOT .......................54
5.4.2 TURNAROUND DELAY INDICATOR AND TURN-AROUND TIME OVERSHOOT ..........................55
5.4.3 SCHEDULE PADDING-GROUND .......................................................................................57
5.4.4 ABSORBED INBOUND DELAY ...........................................................................................59
5.5
SEQUENTIAL ANALYSIS OF REACTIONARY DELAYS ..........................................................64
5.5.1 KEY FACTORS INFLUENCING SEQUENCES OF REACTIONARY DELAYS ..................................64
5.5.2 SEQUENCES IN EUROPE ................................................................................................64
5.5.3 SEQUENCES IN DETAIL ..................................................................................................67
5.6
MAGNITUDE AND DEPTH OF SEQUENCES OF REACTIONARY DELAY ....................................78
5.7
REACTIONARY DELAYS AT EUROPEAN AIRPORTS ............................................................81
5.7.1 REACTIONARY TO PRIMARY DELAY RATIO AT SELECTED AIRPORTS.....................................81
5.7.2 MEAN DAILY IMPACT OF AN AIRPORT ...............................................................................82
5.7.3 AIRPORTS AFFECTING THEMSELVES ...............................................................................86
5.7.4 EXAMPLE OF BAD WEATHER IN FRANKFURT .....................................................................87
CONCLUSION ...................................................................................................................90
OUTLOOK .........................................................................................................................94
8
GLOSSARY .......................................................................................................................96
9
BIBLIOGRAPHY................................................................................................................98
ANNEX 1 : IATA DELAY CODES ...........................................................................................100
ANNEX 2: DESCRIPTION OF CODA DATA...........................................................................103
ANNEX 3: CONVERSION OF UTC TO LOCAL TIME ............................................................104
ANNEX 4: LOW-COST CARRIER DEFINITION......................................................................105
ANNEX 5: AIRCRAFT TYPES AND MEDIAN SEAT CAPACITY ...........................................106
DECLARATION.......................................................................................................................107
LIST OF FIGURES
LIST OF TABLES
INTRODUCTION
This is an empirical study dealing with the propagation of delays in European air
traffic.
1.1
Background
The generally accepted key performance indicator (KPI) for operational air
transport performance is punctuality which can be defined as the proportion of
flights delayed by more than 15 minutes compared to the published schedule.
Other definitions exist, looking at punctuality within 60 minutes of departure/arrival.
35%
Intra-European flights
30%
After
substantial
25%
% of flights
and
20%
15%
10%
5%
Source: AEA*/ CODA
2009
2008
2007
2006
2005
2004
2003
2002
until 2007.
2001*
0%
2000*
(2000-2009)
Due to the high degree of public exposure, it is in an airlines best interest to
operate flights within the commonly accepted 15 minute window. However, there
are many factors that contribute to the punctuality of a flight on which aircraft
operators have no or only limited influence. In reality, punctuality is the endproduct of complex interactions between airlines, airport operators, airport slot
coordinator and air navigation service providers (ANSPs) from the planning and
scheduling phase up to the day of operation.
From a scheduling point of view, which is often months before the day of
operation, the predictability of operation has a major impact to which extent the
use of available resources (aircraft, crew, etc.) can be maximised. The lower the
predictability of operations in the scheduling phase, the more slack time is required
to maintain a satisfactory level of punctuality and hence the higher the strategic
costs to airspace users.
The level of punctuality is closely linked to the level of departure delays. The two
are related to another but the difference needs to be clear. Punctuality allows the
aircraft a 30-minute window around the scheduled time to be on-time or not ontime. Delays, on the other hand, can be positive or negative. Delays are defined as
the time lapse which occurs when a planned event does not happen at the
planned time (Guest 2007: 7). A delay measures the minutes the aircraft is later
or earlier than scheduled. It is the difference between the scheduled and the actual
off-block time for departures, respectively on-block time for arrivals.
On-time performance and delay minutes are key indicators for all stakeholders like
airlines or airports because they are linked to direct costs due to the loss of
productivity as well as to indirect costs due to the invisible loss of time and loyalty
of passengers (Wu 2003b: 418). Mayer (Mayer 2003: 16) states that although
airlines typically blame adverse factors like weather or airport congestion for
occurring delays, there are systematic and predictable patterns to airlines' on-time
performance, meaning that certain delays are foreseeable and handling those
could be implemented in the schedule from the start.
The departure delay of a turnaround aircraft is influenced by the length of
scheduled turnaround time, the arrival punctuality [...] as well as the operational
efficiency of aircraft ground services (Wu 2003a: 329). In conclusion the
Performance Review Unit (PRU) (Performance Review Commission 2008: 32)
stresses that late arrivals originate mainly from late departures. That leads to the
propagation of delays throughout the aircraft rotation and the network of an airline
one delay causing another delay. It is important to note that, for an airline, the
'value' of delay is not just its effect on an individual airframe but its effect on the
operating schedule (Beatty 1998: 2).
Taking a closer look at the different delay causes, the so called 'reactionary'
delays were identified as the largest delay cause (Guest 2007: 29). These
'reactionary', 'knock-on' or also called 'propagated delays' are delays without an
own specific origin or cause. It is the duration of a delay which is transferred from
a previous flight of the same (rotational) or a different (non-rotational) aircraft.
Since generally reactionary delays result from primary delays, they have to be
treated differently and are not to be seen as an individual delay 'cause'.
Even though reactionary delays have a great impact on air traffic performance, the
research effort to better understand and handle them in practice was limited in the
past. Typically primary delays are analysed and taken as main factor for better ontime performance. While critically important due to its contribution to the cost of
delay, it is the primary cause which must be identified if effective action is to be
taken (Guest 2007: 18). However, cost of delay hits airlines twice: both
contingency planning of a schedule (the strategic cost of delay), and then again,
when dealing with the actual delays on the day of operations (the tactical cost of
delay) (Cook 2007: 97). Ahmad Beygi et al. (Ahmad Beygi 2008: 231) confirm the
relevance of reactionary delays: because of the interconnected use of multiple
constrained resources, [...] the propagation of a delay in a flight network has
greater impact than the root delay itself. In CODAs annual DIGEST 2008 (CODA
2009: 34) the impact of reactionary delays becomes apparent, where the share of
reactionary out of all delays account for about 40 percent of total generated delay
minutes.
Overall, the propagation throughout the network is such an inter-related complex
issue, that analysing it, finding patterns, or even trying to predict consequences is
linked to many uncertain variables. Next to qualified information about airlines'
scheduling, fleet and policies, as well as airport congestion and operations,
exogenous factors, for example weather occurrences or in some cases politics,
need to be considered.
In order to minimize the propagation of a delay, airlines can choose a longer
layover on the ground to buffer against the risk of late incoming aircraft or
schedule longer flight time to absorb potential delays on the taxiways (Mayer
2003: 1). Extra time on the ground is cheaper, but accurate anticipation of
[additional time during the block phase] helps with better [] maintenance [and
crew] planning (Cook 2007: 118). In addition to the padding of the schedule,
airlines may have a spare aircraft, flight crew, or ground personnel available.
While these measures decrease the cost of delays when they occur, they also
increase costs of day-to-day operations (Gillen 2000: 3). It is always important to
bear in mind that there is a trade-off between any kind of buffer time and daily
aircraft productivity: the higher the aircraft utility, the higher the revenue. Therefore
a waiting aircraft with unused buffer time includes always sunk costs, because it
can only gain money while flying. Just five minutes of unused buffer, at-gate, for a
B767-300ER, would amount to well over 50.000 over a period of one year, on
just one leg per day" (Cook 2007: 118). 50.000 a year equals to 27,40 a minute.
In Evaluating the true cost to airlines of one minute of airborne or ground delay
the Performance Review Commission (PRC) published also different unit costs.
Passenger delay costs incurred by airlines in consideration of both hard and
soft costs are estimated as 0,30 per average passenger, per average delay
minute, per average delayed flight (University of Westminster 2004: p.51). Based
on their calculations, a delay over 15 minutes has a network average value of 72
per minute (University of Westminster 2004: 100). These costs were adjusted by
inflation to 77 in 2006 (Performance Review Commission 2008: 42). It considers
direct reactionary delay costs, but not the strategic costs through added buffer
minutes. Theoretically strategic buffer minutes should be added to the airlines'
schedule up to the point at which the cost of doing this equals the expected cost of
the tactical delays they are designed to absorb (Guest 2007: 22). The break-even
point was estimated to be a buffer time of the average tactical delay [when] more
than 22% of flights are expected to be delayed by more than 15 minutes
(University of Westminster 2004: 102).
Another and more drastic way of avoiding delays is cancelling flights. This enables
airlines to return to scheduled times and good on-time performance. Nevertheless,
analyses on costs of delays in correlation to network performance in the US
indicated that operational strategies that emphasize maintaining flights even when
there are high delays are more efficient than cancelling flights (Gillen 2000: 13).
For all this, a certain amount of delay is well accepted by the airlines. Following, it
is even more convenient to find out more about the consequences of an occurring
delay, (in a sense of additional costs through rotational and non-rotational knockon delays).
1.2
Objective
The objective of the study is to better understand the processes and mechanisms
of delay propagation in Europe, and to identify factors which amplify or mitigate the
delay propagation.
If an aircraft arrives late at its destination, the delayed inbound flight may not only
be delayed on its next flight leg but it may also affect other flights within the airline
network. This analysis is based on actual flight-by-flight data (and therefore on a
detailed microscopic level) provided by airlines. Through the tracking of aircraft
registrations throughout their rotations, and considerations of different scheduling
strategies of various airlines, the actual propagation of delays is observed and
push factors found.
After a high level analysis of reactionary delays in Europe, more detailed analysis
is carried out to better address the following three issues:
firstly, the delay propagation is analysed from a single airline point of view
by looking at possible differences in airline business models and scheduling
strategies;
The findings can help to improve airline and airport planning in order to achieve a
higher level of resilience towards predictable and unpredictable primary delays.
Furthermore, the findings aim at providing more detailed insights on delay
propagation, which can be useful for macroscopic analyses and simulations.
1.3
Study Scope
For data consistency reasons, the following geographical and temporal scope was
applied.
1.3.1 Geographical scope
The geographical scope of the study is the European Civil Aviation Conference
(EACA) area, as shown in Figure 2. The ECAC area currently consists of 44
Member States comprising almost all European States.
Source: http://www.ecac-ceac.org/index.php?content=lstsmember\&idMenu=1\&idSMenu=10
1.4
Chapter 4 describes the key indicators and the general approach used for
the evaluation of the delay propagation;
Chapter 6 draws conclusions from the results in the previous chapters: and,
LITERATURE REVIEW
Since detailed flight data are commonly available in the US but not in Europe, past
research on reactionary delay considered mainly US air traffic.
For the lack of data it has been very difficult to analyse network effects on a
macroscopic view or detailed aircraft rotation mechanisms on a microscopic view
for European air traffic. Following various papers of previous research are shortly
introduced.
Already in 1998 Beatty, Hsu (both American Airlines), Berry and Rome (both Oak
Ridge National Laboratory) analysed flight propagation through an airline schedule
with the concept of a 'Delay Multiplier'. The delay multiplier is the relation between
the sum of the initial and down line delays, and the initial delay itself. Using
American Airlines data, including crew and aircraft connectivity, they wanted to
develop a 'generic' total value of both the initial delay and its continuing
consequences on the airline schedule (Beatty 1998: 2) Within their concept of the
delay multiplier they considered rotational as well as non-rotational reactionary
delays through crew and passenger connectivity, as well as gate-space limitations.
They found that a linear increase in delay multiplier with increased departure
delay [...] worked well (1998: 5) and that even a small reduction of long root
delays can have a significant affect on total delay in an airline schedule (1998:
7). They concluded, that their results are most probably not valid for different
scheduling strategies, assuming that the delay multiplier would be much smaller
for a large international operator with long turn times and little crew and aircraft
branching [...] while a high frequency, short turn time operator might be much
larger (1998: 8). Finally they analysed the problem of calculating costs due to
cancelling flights and reassigning resources, and suggested to use the cost
calculated by delay multiplier as a conservative surrogate (1998: 8).
Wu published in October 2003 a theoretical study on punctuality performance of
aircraft rotations in a network of airports, analysing different scheduling strategies
in a mathematical model. He observed that the propagation of knock-on delays in
aircraft rotations is found to be significant when short-connection-time policy is
used by an airline at its hub airport (Wu 2003b: 417). When scheduling short
exceptional event alter the predicted delay (2003: 16). Firstly they found out that
the actual flight duration exceeds the planned or announced one, up to 6 minutes
(2003: 9). Additionally, they discovered that short delays between three to fourteen
minutes result mainly from the propagation of a former delay and/or the local
conditions (Load, scheduled stop time) and that propagation and local effects
alone cannot reach values up to 15 minutes if they do not result from an event or
an ATFM slot (2003: 25). For long delays they saw the morning delays absorbed
during long turnaround times by the middle of the day whereas the propagation
and the local effects sustain the level of event or ATFM delays in the evening until
the night stop (2003: 26). Finally, they stated that a flight experiencing a
disruptive event or an ATFM regulation at a station is very likely to undergo a long
delay due to the propagated and local contributors alone, [...] especially [...] during
the latest stations of a daily itinerary (2003: 28).
In 2008 Ahmad Beygi, Cohn, Guan and Belobaba published the Analysis of the
potential for delay propagation in passenger airline network. They investigated the
relationship between schedules and delay propagation with flight data by two
major US airlines, one with mainly hub-and-spoke and one with point-to-point
operations. They examined a delay without looking at other flights at the same
time. Then they created a tree-structure for the following flights of the same aircraft
as well as for the ones which are affected through that single flight. Impacts
through cabin crew and passengers as well as recovery options are excluded.
Throughout their analysis they looked at the sum of propagated delay minutes, the
ratio of the propagated to the root delay, the number of affected flights, number of
flights of the longest propagation sequence, the ratio between the longest
sequence and total number of affected legs, the number of flights where crew
changed aircraft and the ratio of the split up of crew changes to the total number of
affected flights.
They disproved the assumption that a higher number of affected flights correspond
to a higher splitting rate of resources (crews). They also found out that extreme
cases are quite rare (Ahmad Beygi 2008: 224). The maximum count of affected
flights was 7 and 10, for the two operations. About 40 percent of delays of 180
minutes did not propagate at all and about 90 percent had an impact on three or
10
less flights. They observed that delays typically originate at a spoke airport and are
absorbed either at the following or at the second stop at the hub airport. In addition
they stated that the ratio of propagated to root delay decreases as the root delay
occurs later into the day. Also delays benefit earlier in the day more substantially
from increased slack (2008: 232f). The optimal location for the slack is in the
middle of the chain. This is the trade-off point, where the expected delay is
minimized, trading off the lengthy of the propagation and the probability of the root
delay (2008: 236).
In October 2008 Akira Kondo from the Federal Aviation Administration, FAA,
presented the Tail Number Tracking Methodology at INFORM in Washington DC.
As an indicator for propagation performance, a multiplier was calculated by
dividing the arrival delay by the previous arrival delay. Thus, they put the spotlight
specifically at arrival delays. The multiplier is calculated for each leg with a
previous arrival, departure and arrival delay greater than zero. By the end of the
propagation sequence, an overall multiplier as a geometric mean of the single
multiplier evaluates the sequence. Additionally a 'propagation accelerator' is
calculated as the ratio of the propagated delay, which is the minimum of the
previous arrival, departure and arrival delay, and the previous propagated delay.
Finally they presented the ten most affected airports for propagated delay from a
certain airport and to that airport.
Also in 2009, Tony Diana from the FAA published a case study for selected U.S.
airports. He observed that there is no clear evidence that market-concentrated
airports are different from less concentrated ones in terms of delay propagation
(Diana 2009: 280).
According to information from Professor Amadeo R. Odoni and Nikolas Pyrgiotis
from Massachusetts Institute of Technology (MIT), they are developing the
Approximate Network Delays model, AND-concept. It is a macroscopic model
which computes the propagation of delays within a network of airports. The
computation is based on scheduled itineraries of individual aircraft and a queuing
system for each airport. With this tool they want to predict network effects with
different scenarios. So far, they observed that the expected delay relative to
11
schedule increase, and schedule reliability decrease later in the day. Also they
discovered that aircraft flying for the first time to a congested airport late in the day
suffer much less delay. Finally, they assumed that airports affect themselves
significantly within a Hub and Spoke system, returning with the outgoing return.
The EUROCONTROL Central Office for Delay Analysis (CODA) receive
operational flight data from airlines, enriched with additional delay information from
the Central Flow Management Unit, CFMU, covering around 60 percent of all IFR
flights in Europe (see next chapter). Based on these data CODA publish annually
and monthly DIGEST-Reports, a detailed analysis of the actual delay situation in
Europe. However, when looking at changes in traffic flow, year-to-year trends and
delay causes, they concentrate mainly on primary delay causes. In here they
analyse airports, city pairs, and an overall overview. Finally CODA present the
Percentage of all causes Delay by IATA Category (Figure 5), which shows that
the share of reactionary delays sum up for 40 to 45 percent of all delays of all
generated delay minutes.
12
This chapter describes the data sources used for the analysis. It furthermore
describes difficulties and shortcomings related to the data processing and
validation in order to develop a sound basis for the analysis of delay propagation
in Europe in the following chapter of this report.
3.1
Data Sources
Generally, there are many different data sources for the analysis of operational air
transport performance. For consistency reasons, the data in this study were drawn
from a combination of centralised airline reporting and operational Air Traffic
Management systems.
3.1.1 Central Flow Management Unit (CFMU)
In Europe, data are derived from the Enhanced Tactical Flow Management
System (ETFMS) of the Central Flow Management Unit (CFMU) located in
Brussels, Belgium.
The system stores data repositories with detailed data on individual flight plans
and tracks sample points from actual flight trajectories. It enables CFMU to track
Air Traffic Flow Management (ATFM) delays by airport and en route reference
location.
3.1.2 Central Route Charges Office (CRCO)
The second, centralised data collection comes from the Central Route Charges
Office (CRCO). As the name states, they calculate the fee for the air space use of
a state, invoice it to the airlines and reimburse the states, respectively the ANSPs.
For this purpose, the CRCO uses an efficient cost-recovery system that funds air
navigation facilities and services and supports Air Traffic Management
developments (CRCO homepage).
3.1.3 Central Office for Delay Analysis (CODA)
CODA aims to provide policy makers and managers of the ECAC Air Transport
System with timely, consistent and comprehensive information on the air traffic
13
delay situation in Europe, and to make these available to anyone with an interest
in delay performance (CODA homepage).
In Europe, CODA collects data from more than hundred airlines each month. The
data collection started in 2002 and the reporting is voluntary. Currently, CODA
covers 60 percent of all IFR flights in the European Civil Aviation Conference
(ECAC) area which includes 44 countries. Figure 3 illustrates the coverage of
CODA data by ECAC Member State for July 2009. For instance, the data
submitted by airlines in July 2009 covers 69 percent of all German IFR-departures.
Source: CODA
Out of the gate, Off the runway, On the runway, and Into the gate.
14
submitted by the airlines it is possible to evaluate the propagation of delay and the
underlying causes. Airlines may use up to delay 5 codes per flight to specify the
reason of a departure delay. This valuable information is neither gathered by
CFMU nor by CRCO. Usually, when working with calculated propagated delay
minutes, there is always one open question: which part of the departure delay is
newly added and which is propagated from a previous flight leg? The regular flight
data give no information about which part of the inbound delay was absorbed and
how much propagated to the next flight.
With the reported delay codes the actual reactionary delay and the new primary
delay can be separated from each other. They demonstrate how much of the
inbound delay has been absorbed, thus, how long the delay would have been if
there was no primary delay. Figure 4 illustrates this significant advantage of CODA
data in more detail:
the aircraft at the top of Figure 4 operates according to the scheduled turnaround time with no additional delays.
00:10
Turnaround
Turnaround
00:20
00:30
00:40
new
primary
delay
taxi out
Inbound
delay
10 min.
absorbed
inbound
delay
reactionary
delay
taxi out
Inbound
delay
10 min
00:00
taxi out
taxi in
taxi in
Turnaround
taxi in
with primary
delay
Turn-around..
as scheduled
with
reactionary
delay
00:50
01:00
Turn-around time
15
Without the delay codes reported by the airlines, it would not be possible to
differentiate between the new primary delay (middle bar of Figure 4) and the
reactionary delay (bottom bar of Figure 4).
3.1.3.1 IATA Delay Coding
In order to foster the harmonised reporting of delay among its member airlines,
The International Air Transport Association (IATA) has published a standard
coding system for the classification of delays (see Annex I).
As shown in Table 1, the IATA delay codes can be broadly divided into ten parts,
according to the area of accountability:
Table 1: Standard IATA delays codes
IATA Code
Definition
0-9
11-18
21-29
31-39
41-48
51-58
61-69
71-77
81-89
91-96
97-99
16
to
the
better
Definition
91
92
93
94
95
96
17
IATA delay code 93 is also called rotational reactionary delay, because it relates
to reactionary delays on successive flight legs of the same aircraft (Figure 6).
Aircraft 1
1
Code 93
Aircraft 2
Codes 91, 92
94, 95, 96
Awaiting crew,
connecting passenger, etc.
Primary delay
18
94-95
96 91
92
93
Figure 7: Split-up of
reactionary delays
By far the main share of reactionary delay is due to rotational reactionary delay
which accounts for 89 percent of all reactionary delay reported during the analysed
period.
Figure 8 shows the distribution of the reactionary delay categories by airline
business model (low-cost, hub-and-spoke, and point-to-point) and by time of the
day in summer 2008. Morning lasts from 6:00h till 13:59h, Afternoon from 14:00h
till 21:59h and Night from 22:00h till 5:59h.
19
Irrespective of the airline business model, the rotational delay accounts by far for
the highest share of reactionary delay.
Hub-and-spoke operations show with 15 percent the largest share of nonrotational reactionary delay which is normal in view of the type of operations.
As can be expected, low-cost and pointto-point operations only show a small
share of non-rotational reactionary delay as they often operate independent
services without the need to wait for connecting passengers or load. The nonrotational delay reported by those carrier types is mostly related to crew (code
94/95) or operations control (code 96)
Irrespective of the type of operations, the main share of reactionary delay (around
60 percent) is reported in the afternoon, followed by the morning (25-30 percent)
and the smallest share during night.
For all three business models codes 91, 92 and 94-95 are higher in the afternoon
than during morning or night time. This indicates that airline focus in the afternoon
is more on managing flight connections while in the morning the focus is more on
schedule adherence.
3.2
The most important prerequisite of the delay propagation study is the data
processing and validation in order to develop a sound basis for the analysis in the
next chapters of the report.
A considerable amount of time was necessary to prepare the vast amount of data
available from the various data sources (see section 3.1) and to resolve
inconsistencies in order to develop a data base for the analysis of delay
propagation.
This section describes the encountered difficulties in the processing of the data
and the applied solutions.
3.2.1 Missing or incomplete data
One of the most influencing limitations for the analysis of delay propagation is
missing or incomplete data. For the development of rotation sequences of an
20
aircraft the exclusion of only one flight due to missing or incomplete data means
that the entire rotation sequence is incomplete.
For the analysis of delay propagation, the aircraft registration is one of the key
parameters needed in order to create the rotation sequences. If key parameters
are missing, the data submitted by airlines is cross-validated with data from the
CFMU or CRCO in order to complete the missing data (see Figure 9). In cases
where the aircraft registration could not be retrieved from one of the three sources
the respective data record was rejected from the analysis.
CFMU
CODA
Data sample
for reactionary
delay analysis
CRCO
21
mandatory and therefore some airlines have developed their own or slightly
modified delay coding schemes in order to meet their operational needs.
For comparability reasons, CODA has developed algorithms to recode tailored
coding schemes into the standardised IATA coding scheme.
Only flight data using the standardised IATA delay coding or for which the data
could be recoded is included in the analysis in the next chapters of the report.
3.2.3 Different coding policies
Note that the reported delays are based on a persons decision2 and therefore are
to some extent subject to interpretation and airline preferences. A different person
could report the same delay differently. However, most airlines have a specific rule
for exactly that issue and knowing how these strategies work, helps dealing with
the reported codes.
Generally airlines aim at reporting the delays as they occur, but what if there are
two reasons at the same time (i.e. ATFM delay AND boarding delay), or it is simply
not known what really caused the delayed minutes? Airlines split the delays
according to their respective duration, but some only report the most penalising
or longest delay, others split the minutes in half and report both. Reporting the
cause of the longest delay reduces the visibility of shorter primary delays for the
respective airline (i.e. 5 min. delay due to boarding will be hidden behind a 20
minute ATFM delay). Reporting both delays would, on the other hand, reduce the
visible impact of the longer delay.
One major carrier in Europe reports not the longest, but always the last delay
cause. For instance when an aircraft with 30 minutes of reactionary delay gets an
additional small delay all previous delay minutes are reassigned to the new
primary delay cause. If this practice is not known, one would assume that the
aircraft recovered from all previous delay minutes, and that the new delay had a
bigger impact than it really had.
The delay codes are often given by the handling agend and are then sent to the
Operational Control Centre (OCC) of the airline with the aircraft movement message
(MVT).
22
Another reporting practice particularly relevant for the analysis of reactionary delay
was observed for another airline. The airline exclusively uses primary delays even
if they are carried over as reactionary delays on the next flight leg. The advantage
of this technique is that the actual root cause remains visible on the subsequent
flight. For the analysis it is however impossible to identify whether it was a
reactionary delay or another primary delay.
Another difficulty in reporting reactionary delays is to separate them from primary
delays on the ground. In practice, it is not always obvious how many minutes were
transferred from a late arrival and how many minutes were added newly. In other
words, it needs to be decided if there was extra buffer time considered in ground
phase, and whether it was used only for the inbound delay or also for parts of an
eventually new delay. When it is not clear, many airlines just split the delay in half
and report both, a reactionary and a new delay.
In order to avoid any bias from different coding practices, airlines applying a
coding practice which could spoil the analysis were excluded from the analysis.
The results of the analysis should nevertheless be viewed with a note of caution
due to possible differences in the interpretation of IATA delay codes.
3.2.4 Errors in datasets
During aircraft registration tracking sometimes different aircraft types showed up
for one aircraft registration. As aircraft registrations are unique for every aircraft,
each aircraft can only have one aircraft type. In order to resolve this issue, the
actual aircraft type was determined by analysing the frequency of occurrence in
the data during the analysed period. The aircraft type for the few records with a
different aircraft type was then aligned with the most frequent type for this aircraft
registration.
The aircraft type is used in the analysis to group each aircraft type according to its
median seat capacity3. A more detailed table with aircraft types and their median
seat capacity can be found in Annex V.
23
When calculating the block times of flights another error occurred quite frequently.
For about 170 datasets the block time exceeded 1440 minutes, which is equal to
one day. Apparently in those cases the date of arrival was simply put falsely one
day after the departure date. The data was cleaned by manually correcting the
date.
Another problem was linked to arrival times. The calculation of automatically
computed fields, like the arrival delay could for example not handle flights with an
ON or IN-time at midnight. Therefore arrival, landing, and taxi-in delays were
recalculated in those cases.
3.2.5 Missing flights
For the development of aircraft sequences the exclusion of only one flight means
that the entire rotation is incomplete. For instance, some smaller non European
airlines only report the flights bound for Europe.
Two regional airlines only send information about delayed flights but not about the
flights which were on-time.
As it is impossible to build aircraft sequences when flights are missing, those
airlines which only report a part of their flights were excluded from the analysis.
3.3
Input in analysis
24
Traditional
scheduled
Low-cost
Cleaned data of
flights of
selected
airlines
Flights in
complete
rotations
6.527.962
95,6 %
73,8 %
2.116.704
96,0 %
82,4 %
The last column in Table 3 represents the sample used for the analysis in the next
chapters. It only includes sequences with more than one rotation and only panEuropean flights.
Overall, about 50 percent of all IFR departures in the ECAC area are included in
the sample used for the analysis of delay propagation.
3.4
Data Processing
In view of the vast amount of data that needs to be processed for the analysis, the
Statistical Analysis System (SAS) Enterprise Guide was used for the entire
analysis. After 'cleaning' the airline data, as described in the previous section, this
section describes how rotation sequences of aircraft are build for the further
analysis in Chapter 5.
3.4.1 Building sequences with airline rotations
Rotational sequences are built by linking individual flights through their unique
aircraft registration over time. This also serves as the final control to ensure that
the sequence is complete.
In order to create these sequences, all flights are grouped according to their
unique aircraft registration and sorted by their actual reported off block times
(Figure 10).
In a next step, the individual flight legs for each unique aircraft registration are
connected by date and time and by their ICAO airport designator. For example,
the arrival airport on one flight leg has to match with the departure airport on the
25
subsequent flight leg and so on. As the sample relates to pan-European flights,
most sequences start in the early morning and end at night.
1 sequence
with
3 rotations
Single flight
perspective
Sequential
perspective
The SGT limits for each of the four groups are based on median seat capacity,
average observed SGT and expert judgement from EUROCONTROL staff working
in this area.
The SGT is the difference between the scheduled arrival time (STA) of the previous flight
and the scheduled departure time (STD) of the subsequent flight.
26
There is only one exception to the SGT limits outlined in Table 4. When a flight
exceeded the SGT limit but reported a reactionary delay on the next flight leg, the
flight was kept in the sample.
3.4.2 Grouping by airline business model
The following additional attributes and groupings were applied in this study.
As airline business models are expected to react differently to the propagation of
delay, the flights were categorised according to three different business models:
Point-to-point operations.
Airport
A:
START
Airport
A
Airport
B
HUB:
START
Airport
C
Airport
C
Airport
B
Hub-and-spoke System
BASE
Point-to-point system
27
28
29
CONCEPTUAL FRAMEWORK
After the description of data processing and validation issues in the previous
chapter, this chapter presents the approach of how propagation of delays is
analysed.
4.1
Primary
delays
(length, time of
day, root cause,
etc.)
Actual operations
This section provides an overview of the key performance indicators (KPI) which
are used to measure and describe the propagation of delays. The sensitivity to
primary delays, scheduling tactics, sequences of reactionary delays, as well as
reactionary delays from an airport point of view are analysed. The analysis
differentiated between the three different airline business models described in the
previous chapter.
30
Ratio
reactionar y delay
primary delay
The higher the ratio, the more sensitive is the operational system to primary delays
and the more reactionary delay minutes are generated for each minute of primary
delay. This draws a high-level picture of the impact and the importance of
reactionary delays in European air traffic and reveals differences between the
business models.
4.2.2 Airline scheduling matters
Scheduled turn-around and block-to-block times play an important role in
absorbing and reducing primary and subsequent reactionary delays.
Block time
Block time
Turn around
Block time
Turn around
Actual IN
Actual OUT
STD
31
In this study, two complementary indicators were used to get a first understanding
of performance differences in the block-to-block phase.
I. Delay Difference Indicator Flight (DDI-F)
The DDI-F provides an order of magnitude of the deviation between the scheduled
block time and the actual flown block times. It is expressed as mean deviation
compared to the scheduled block time.
[min]
[%]
32
Similar to the DDI-F and the BTO, two complementary indicators are used to get a
first understanding of whether an airline is able to stick to its scheduled ground
times.
I. Delay Difference Indicator Ground (DDI-G)
The DDI-G provides an order of magnitude of the deviation between the scheduled
turn-around time and the actual observed turn-around times. It is expressed as
mean deviation compared to the scheduled turn-around time.
DDI - G departure delay - inbound delay
[%]
It is important to note that the DDI-G can include additional time in both directions
of the schedules ground time: early arrivals are considered just like added delay
during turn-around.
In order to take a deeper look at the turn-around process itself, two other related
and again complementary indicator are found to be more useful in terms of delay
analysis during the ground time.
III. Turn-around Delay Indicator (TDI)
The TDI equals the DDI-G but neutralizes early arrivals. The actual arrival time is
set to the scheduled arrival in case of an early arrival.
33
Inbound delay 0
Inbound delay 0
THEN
THEN
Therefore this indicator shows the general tendency of an airline to absorb or add
delay during all ground phases.
IV. Turn-around Time Overshoot (TTO)
Similar to the GTO, the TTO indicates the percentage of flights, which still outrun
the scheduled ground time when early arrivals are neutralized.
TTO
The TDI and TTO demonstrate whether and how much delay is added in general
during the ground time.
However, for the analysis of the propagation of delays, the reaction following an
inbound delay needs to be looked at individually. There the schedule paddingGround is introduced as another IF-indicator of the DDI-G.
V. Schedule padding-Ground (sched.pad-G)
The schedule padding-Ground measures the deviation of the actual to the
scheduled ground time, IF the aircraft arrived late. It seems similar to the TDI, but
reveals slightly different information.
34
4.3
Absorbed delay
Inbound delay
Reactionary delay
35
During the first ground phase, the aircraft absorbs part of the inbound delay, but
suffers from a new primary delay. Therefore it departs with the remainder of the
inbound delay as reactionary delay and the new primary delay.
In the second block-to-block phase, the aircraft recovers a little and arrives with
less inbound delay than it had on departure. In the second ground phase, the
aircraft is able to absorb a big part of the inbound delay and departs finally with
little reactionary delay, consisting of primary delays from the two previous flight
legs.
4.3.2 Root delay
An important role in a sequence of flights with reactionary delays is the 'root delay'.
The root delay is the first primary delay which caused a reactionary delay in a
sequence. The root delay can be due to:
The root delays are grouped according to their duration into the following groups:
1 to 15 minutes
16 to 60 minutes
61 to 120 minutes
36
Magnitude
An important issue to point for the interpretation of the multiplier or the magnitude
is its sensitivity to the length of the root delay and to the length of the sequence of
reactionary delays.
For example, if a 10 minute delay results in a 10 minute reactionary delay, the
multiplier equals 1 (10/10=1). If a flight has a root delay of 60 which causes a
reactionary delay of also 10 minutes, the multiplier is 0,17 (10/60=0,17).
The same amount of reactionary delay results in a different multiplier depending
on the length of the root delay. Hence, for the comparison of delay multipliers it is
necessary to take the length of the root delay into account.
Similarly, the multiplier also depends on the number of flight legs affected by
reactionary delay.
37
This chapter provides an analysis of the reactionary delay in Europe. The chapter
consists of the following complementary parts:
section 5.2 analyses the sensitivity of the three airline business models to
primary delays. The ratio between the reported reactionary and primary
delay is taken to get a first understanding of fundamental differences
between business models and over time (see 4.2.1);
section 5.3 looks at differences in the scheduling of the block phase among
the three airline business models and the respective impact on the
propagation of reactionary delay (see 4.2.2.1);
38
5.1
Summer 2008
1-15 min.
121-180 min
50
40
30
20
10
0
by nr of flights
delay
in
terms
of
61-120 min
60
% of observations
of reactionary delays.
16-60 min
>180 min
by primary delay
minutes
occurrence.
However in terms of minutes generated, primary delays from 16 to 60 minutes
account for the highest share.
5.2
As described in 4.2.1 the ratio between reactionary and primary delays is used as
key performance indicator for the sensitivity of the three airline business models to
reactionary delays.
The ratio describes the impact of primary delays on successive flights in form of
reactionary delays. If the ratio is one, every minute of primary delay generates an
additional minute of reactionary delay. If the ratio is higher than one, the total
amount of reactionary delay generated is higher than the amount of initial primary
delays.
5.2.1 Methods of calculating reactionary delay
As outlined in the literature review in Chapter 2 due to the lack of available data most studies on reactionary delay are based on models which calculate the delay
propagation based on available airline schedules. The advantage of this
methodology is that the assumptions made for the calculation are not influenced
39
Figure 18 compares the results of the two different methodologies. The solid lines
indicate the mean ratio of the (reported) reactionary delays of the different airline
business models, the dotted line the ratio of the (calculated) propagated delays.
With some exceptions in specific months, the correlation between the (calculated)
propagated delay and the (reported) reactionary delay is generally good for all
three airline business types. It is interesting to note that the reported delay (solid
lines) is most of the time below the calculated delay (dotted lines). This is most
likely due to the inability of the model to account for a certain level of flexibility to
speed up turn-around times when required.
The analysis in the following sections only relates to reported reactionary delay.
40
41
The
ADDD
for
point-to-point
higher
share
of
were
and
delayed
low-cost
during
summer 2008.
delayed
departures [%]
comparatively
30
25
20
15
10
5
51 %
46 %
39 %
80
Hub-and-spoke
Low-cost
59 %
59 %
Point-to-point
60
40
20
44 %
provides a first high level estimate of Figure 19: Share of reactionary delay
the importance of reactionary delay for by type of operation (Summer 2008)
each of the three airline business
models.
The analysis in the next section of this report provides an analysis of the
reactionary to primary delay ratio over time for the three airline models.
42
of
operation.
Two
peaks
are
resulting
in
significant
Hub-and-spoke operations
operations,
show
the
operations
improvement
show
since
a
the
beginning of 2009.
Low-cost operations have the highest
ADDD (26 minutes) and also the highest
share of reactionary delay (13 minutes).
On average, every minute of primary
delay generated more than one minute of
reactionary delay for this type of business
model.
43
Point-to-point operations
Ratio of reactionary to primary delays
-- Primary delay
-- Reactionary delay
Figure 20: Seasonal evolution of
reactionary delay ratio
The observations in Figure 20 are consistent with the findings in Figure 19 Low-cost
carriers show on average the highest level of departure delay (primary +
reactionary delay) and hence the highest ratio. The 51 percent accounting for
reactionary delay in Figure 19 equal a ratio slightly above one.
5.2.2.2 Ratio reactionary/primary delay within the week
Figure 21 and Figure 22 show the within-week-variation of the reactionary to
primary delay ratio. The ratio is furthermore put into context with the number of
flights (Figure 21) and the average delay of delayed departures (Figure 22).
The ratios of hub-and-spoke and point-to-point operations show a similar pattern
with a clear peak on Fridays (5) and the lowest level on the weekends. Low-cost
operations, on the other hand show a slight drop on Wednesdays (3) and
Thursdays (4).
44
Figure 21 suggests that the ratio is not directly linked to the level of traffic. For
instance, point-to-point operations have the lowest mean number of flights but the
highest reactionary to primary delay ratio.
45
The ratios of all three types of operation show a similar pattern throughout the day
of operations.
As most European airports have night flight restrictions and traffic demand is
limited during night-time, only a limited number of flights are operated during night
time. While the ratio is very high during the night, the number of flights is very low
and the results should therefore be viewed with a not of caution. The ratio is to
some extent artificially high because delayed departures consist mainly of
propagated delay minutes which accumulated throughout the day and which
therefore strongly impact the calculation of the ratio of the few flights still
operating.
Airlines usually schedule their first flight of the day in the early morning. All three
types of operations show a traffic peak in the morning at around 7:00h and a
second peak in the afternoon at around 16:00h. The delay ratio increases
continuously after each traffic peak and shows only a decrease in the early
afternoon when traffic levels are reduced.
46
47
The relation, between the ADDD and the mean reactionary delay of delayed
departure, is shown in Figure 25. Clearly the gradient is below one, so that the
mean reactionary delay does not increase as much as the mean departure delay.
The trend can be caused by two factors: firstly not all primary delays propagate.
Secondly, the impact of reactionary delays is lower than the impact of all primary
delays together. This was already observed in Figure 5 and Figure 19 where the
share of reactionary delay accounted for some 40 percent of all reported delay.
However, an expected increase in form of an exponential distribution with a higher
ADDD cannot be observed.
Winter 2007-08 to summer 2009
40
35
30
25
20
15
10
5
0
0
10
15
20
25
30
35
40
ADDD [min]
Low-cost airlines
Hub-and-spoke airlines
point-to-point airlines
Figure 25: Average delay and reactionary delay per delayed departure
48
5.3
In this section block phase performance of the airlines is analysed with the key
performance indicators detailed in 4.2.2.1: Delay Difference Indicator-Flight (DDIF) and Block Time Overshoot (BTO). It is important to point out that the DDI-F is
the mean absolute difference between the actual and the scheduled block time
and the BTO is the percentage of flights exceeding the scheduled block phase.
From a scheduling point of view, the predictability of operations months before the
actual day of operations has a major impact on the utilisation of available
resources (aircraft, crew, etc.). The lower the predictability of the necessary blockto-block time, the more time buffer is usually required to maintain a satisfactory
level of punctuality. The level of schedule padding is subject to airline policy and
depends on the targeted level of on-time performance and notable differences
between airlines can be observed.
When looking at scheduled block times or departure and arrival times, marketing
strategies and airport slot allocation needs to be considered.
Figure 26 shows the relation between the BTO (horizontal axis) and the DDI-F
(vertical axis) by airline and business model for the summer season of 2008.
Summer 08
8,0
6,0
DDI-F [min]
4,0
2,0
0,0
-2,0
-4,0
-6,0
-8,0
-10,0
0
20
40
60
80
100
BTO [%]
Low-cost airlines
Hub-and-spoke airlines
Point-to-point airlines
49
A clear correlation between the two can be observed. With a higher mean DDI-F,
the number of flights exceeding the scheduled block time rises as well. In the
graph each dot represents one airline.
Operational performance in the block-to-block phase varies among the analysed
airlines, as the range of the mean DDI-F spans from minus eight to plus four
minutes. Overall, the mean DDI-F is slightly negative for all business models. Lowcost airlines absorb on average some five minutes per flight, hub-and-spoke
airlines three minutes and point-to-point airlines only two minutes. However, they
all plan generally more block time than actually required, indicating that a certain
level of buffer time is included in the scheduled block-to-block times.
Note, that the inclusion of buffer time in the block-to-block phase (DDI-F < 0) has
also disadvantages as it reduces aircraft and crew efficiency. Crews are assigned
to fly longer block times than they really do, which leads to additional costs due to
slack time in the crew scheduling.
Apart from this negative impact on airline efficiency, time buffers in scheduled
block-to-block times result in a certain level of aircraft to arrive ahead of their
scheduled times (early arrivals ). This in turn may have an impact on airport
operations as facilities and stands may not be readily available.
On average, there are only a few airlines which actually generate delay as a result
of insufficient scheduled block-to-block times.
Within the three operations, low-cost operators have with around minus five
minutes the lowest DDI-F. Only one of the observed low-cost airlines has a
positive mean DDI-F, during one season its DDI-F jumps up to even eight minutes.
On average, between 70 percent and 85 percent of this specific airlines flights
exceed the scheduled block time (see also green dot in upper right corner of
Figure 26). All other airlines with low-cost operations plan one to even ten minutes
more than they actually need, thus, having a scheduled block time ten minutes
longer than the actual block time. Most these airlines see 20 to 30 percent of
flights exceeding the scheduled block time.
50
As can be seen in Figure 26, hub-and-spoke operators (red dots) have generally a
slightly negative DDI-F of around minus three minutes and a BTO between 30 and
40 percent. However the picture is contrasted among hub-and-spoke operators.
While one hub-and-spoke carrier shows for example a comparatively low DDI-F of
around minus 7,5 minutes and a BTO of only 17 percent (see red dot in the bottom
left corner of Figure 26) another carrier exceeded the scheduled block time by 3
minutes on average and up to 75 percent of the flights exceeded the scheduled
block time.
Because of the low DDI-F, the first hub-and-spoke carrier is able to recover at
least one third of its flights which departed with a departure delay during the blockto-block phase. However, this also implies that a comparatively high number of
aircraft arrive even before their scheduled arrival time, in this case, between 55
and 70 percent.
The huband-spoke carrier (with the positive DDI-F) generates on average already
delay during the block-to-block phase. Consequently the share of delayed flights
increased by eight percent in the block-to-block phase. During the summer season
2008, 76 percent of arrivals were delayed with a mean arrival delay of 35 minutes.
Logically this leads to an increased probability of reactionary delays.
The impact of the DDI-F in terms of number of delayed flights is visualized in
Figure 27. It illustrates, that carrier with no or positive DDI-F are likely to increase
the number of delayed flights on arrival, or even double them, during the first block
phase.
Flights of airlines absorbing at least two minutes during the first block phase are
able to reduce the number of delayed flights.
51
Summer 2008
Nr of delayed arrivals / Nr of delayed
departures on first leg
3,0
2,5
2,0
1,5
1,0
0,5
0,0
-10
-8
-6
-4
-2
0
DDI-F on first leg [min]
Low-cost airlines
Hub-and-spoke airlines
Point-to-point airlines
The following section evaluates the relation between inbound delays upon arrival
and reactionary delays on the subsequent flight leg. The inbound delay is the
observed delay when the aircraft arrives at its destination airport. Depending on
the performance during the block-to-block phase, the inbound delay can be larger,
smaller or equal to the departure delay observed at the origin airport. It should be
noted that inbound delays are only calculated for flights with a subsequent
departure. Consequently the delay upon arrival on the last flight leg is not
considered in the calculation of the average inbound delay which leads to a lower
average delay than is observed for the average delay per delayed arrival (which
includes the last flight leg)
Figure 28 illustrates the relation between the average delay of delayed inbounds
(ADDI) and the mean reactionary delay on subsequent flight legs. The horizontal
axis represents the ADDI and the size of the bubble represents the percentage of
delayed inbound flights. The vertical axis shows the mean reactionary delay on the
subsequent flight leg.
52
Summer 2008
50
67
40
30
67
20
48
42
74
49
76
37
10
77
32
47
52
63
0
0,0
10,0
65
20,0
30,0
40,0
50,0
Point-to-point airlines
Low-cost airlines
53
5.4
This section evaluates the ability to absorb delays in the turn-around (ground)
phase. The KPIs are described in section 4.2.2.2. The turn around phase in this
study is defined as the time between the IN-time (on-block) and the OUT-time (offblock).
As outlined in 4.2.2, delays can be absorbed in the block-to-block phase and in the
turn-round phase. Since many different players are involved in the turn-around
process, a good planning and a high level of predictability is essential for turnaround efficiency and good performance. Turn-around performance and the ability
to absorb delay during this phase plays therefore an important role in the analysis
of delay propagation.
5.4.1 Delay Difference Indicator-Ground and Ground Time Overshoot
The GTO (percentage of flights exceeding the scheduled ground time) and the
DDI-G (mean actual absolute minutes difference to scheduled ground phase)
describes the ground phase like the DDI-F and the BTO the block phase. Figure
29 shows the relation between the DDI-G and the GTO.
Summer 2008
12,0
10,0
DDI-G [min]
8,0
6,0
4,0
2,0
0,0
-2,0
-4,0
0
10
20
30
40
50
60
70
80
90
GTO [%]
Low-cost airline
Hub-and-spoke airline
point-to-point airline
54
100
For almost all airlines the same relation is observed: the higher the DDI-G, the
more flights stay longer on the ground than scheduled. For low cost operations
there were three clear outliers and the trend line for low-cost operators was
removed in order to avoid confusion. It is striking that irrespective of the type of
airline operations between 60 and 90 percent of all flights have a turn-around time
longer than actually scheduled.
Around 70 percent of the hub-and-spoke operations exceed the scheduled ground
time, leading to a mean DDI-G of almost plus five minutes. Point-to-point operators
have the smallest DDI-G of plus three to four minutes. Low-cost operations tend to
stay even longer and show a DDI-G of plus eight minutes.
5.4.2 Turnaround Delay Indicator and Turn-around Time Overshoot
Figure 30 depicts the turn-around delay indicator (TDI) and the turn-around time
overshoot (TTO) as described in section 4.2.2.2. It is important to recall that the
TDI sets all early arrivals to the scheduled arrival time in order to take out this bias.
Summer 2008
12,0
10,0
TDI [min]
8,0
6,0
4,0
2,0
0,0
-2,0
-4,0
0
10
20
30
40
50
60
70
80
90
TTO [%]
Low-cost airline
Hub-and-spoke airline
point-to-point airline
55
100
In comparison to Figure 29, Figure 30 has shifted to the left and a slightly
downwards. The slope is flatter. This reveals information on early-arrival practices
of airlines as well. Only between 10 and 50 percent of aircraft exceed the ground
time because of new primary delay, whereas almost twice as many flights
exceeded the scheduled turn around time because of a combination of an early
arrival time with a late departure.
Note that the three low-cost carriers with the considerably high DDI-G in Figure 29
align themselves in Figure 30 along the other low-cost carrier.
On average, low-costs operations show the shortest scheduled turn-around time.
Within the given sample they reach a mean scheduled ground time of 40 minutes.
Point-to-point operations are scheduled on average 4 minutes longer than low cost
operations and hub-and-spoke operations are scheduled on average 10 minutes
longer.
However, it is important to point out that the mean turn-around time depends also
on the mix of the aircraft fleet, which limits the ability to directly compare turnaround times. However, it is possible to conclude that the turn-around times of
low-cost operations are scheduled quite tightly.
Consequently it is not surprising to see in Figure 30 that on average low cost
operators exceeded their scheduled turn-around times more often that the other
types of operation. Up to 46 percent of the analysed low cost operations exceeded
their scheduled turn around phase and consequently generated delays. On
average, low cost carriers added four minutes of delay in the turn around phase in
summer 2008.
In comparison, hub-and-spoke operations added only around 1 minute during the
turn-around phase in summer 2008.
The picture is different for point-to-point operations. On average aircraft required
less turn-around time than originally scheduled. The mean TDI was around minus
one minute in summer 2008 and only 25 percent of flights exceeded the scheduled
turn-around time.
56
57
Summer 2008
10
14
25
15
39
13
20
14
10
17
2
-2
11
23
11
27
15
12
18
11
7 8
12
9
-4
12
25
15
-6
0
10
15
20
25
30
35
40
45
50
55
60
58
When looking closely at the values, it becomes apparent that the schedule
padding-Ground actually does not directly link the ADDI to the average reactionary
delay of delayed departures. The reason for this is that the schedule paddingGround does not indicate whether the scheduled turn around time would have
been sufficient in the first place.
5.4.4 Absorbed inbound delay
The last indicator described in 4.2.2.2 is the absorbed inbound delay. With the
delay codes reported by the airlines, the actually propagated reactionary delay can
be identified which enables to quantify the absorbed inbound delay.
Figure 32 shows the relation between the average delay per delayed inbound
(ADDI) on the horizontal axis, the absorbed inbound delay on the vertical axis and
the reactionary delay on the subsequent flight leg (size of the bubbles).
Summer 2008
-1
25
14
15
-3
8
-5
-7
-9
-11
10
17
39
13
14
7 11
8 11
11 14
20
23
18
25
27
15
-13
12
-15
0
10
15 20 25 30 35 40 45 50
Average Delay of Delayed Inbound [min]
Low-cost airlines
Hub-and-spoke airlines
55
60
point-to-point airlines
59
the shorter absorbed delay, the higher is the mean reactionary delay on the
subsequent flight leg.
Compared to Figure 31, in which point-to-point operations show the highest ability
to reduce inbound delay, Figure 32 shows that during the turn-around phase pointto-point operators absorb in reality only about as much as hub-and-spoke carriers.
This is consistent with the higher ratio of reactionary to primary delay of point-topoint operations (see chapter 5.2.2) and it leads to the conclusion that point-topoint operations do not suffer as much primary delay during the turn around phase
as hub-and-spoke operators. Therefore, the effect of reactionary delay is higher
which consequently increases the ratio.
Furthermore Figure 32 confirms that low-cost carriers have by far the highest
ADDI, but absorb the least inbound delay during the ground phase (maximum
seven minutes). This leads inevitably to higher mean reactionary delays on the
subsequent flight leg.
Allusively, the graph provides information about the turn-around performance in
terms of additional aircraft suffering primary delay. For most of the airlines the
mean reactionary delay of delayed departures is less than the difference between
the ADDI and the absorbed delay. This is due to the number of aircraft which were
not delayed on arrival but added delay during the turn-around phase. They impact
the average delay of delayed departures but not the average delay of delayed
inbounds.
As an example the airline, represented by the blue bubble on the bottom of Figure
32 is described: The airline has an ADDI of 29 minutes and was able to absorb the
inbound delay by 13 minutes on average. In theory this would result in a
reactionary delay of 16 minutes on the subsequent flight leg if the number of
delayed aircraft stayed the same. However, the mean reactionary delay of the
example airline shows only 12 minutes (instead of 16), which suggests that the
number of delayed aircraft on departure has increased in comparison to the
number of flights delayed upon arrival. More aircraft departing with only primary
delay obviously lower the mean reactionary delay of delayed departures.
60
The following section provided a more detailed illustration of the concept by using
the aforementioned low-cost airline which showed a somehow different behaviour
than the other low-cost airlines.
The low-cost carrier outlined already in section 5.4.3 showed the following
results for the analysed performance indicators:
PDI = 67 %
In Figure 32 this airline is found as a green bubble with almost seven minutes
of absorbed inbound delay (and still 25 minutes of mean reactionary delay).
The difference between the ADDI and the absorbed inbound delay is 25
minutes and the schedule padding-Ground is almost equal to the absorbed
inbound delay. This indicates two things: First, the airline mostly does not add
new primary delay during the turn-around phase and secondly, more aircraft of
this airline depart early than of other airlines.
The second factor is supported by the near zero value of the DDI-G. Also the
comparison of its overall mean departure delay and its overall mean
reactionary delay reveals that 13 of the 15 minute departure delay (87 percent)
are due to reactionary delay. When looking only at delayed departures, 75
percent of the departure delay is due to reactionary delay. Both values are
especially in comparison to the other operators are quite high and confirm
that the airline does not add a lot of new delay during the turn-around phase.
The following section compares now this low cost airline to one of the more
typical low-cost carriers:
PDI = 53 %
The second low cost carrier absorbs nearly six minutes during the block phase.
Only 23 percent of the flights exceed the scheduled block-to-block time.
61
However, 53 percent of the aircraft arrive with a mean delay of 25 minutes per
delayed inbound. Of these 25 minutes only three minutes get absorbed during
the turn-around phase. As the mean reported reactionary delay of delayed
departures accounts only for 15 minutes, a 7-minute gap needs to be
explained. The gap is due to the addition of new primary delay during the turnaround phase. This is confirmed by the share of reactionary delay within the
total departure delay. 15 of the 26 minutes of departure delay per delayed
departure (56 percent) are caused by reactionary delay. Also the high
sched.pad.-G confirms that even delayed flights exceed the ground time on
average by seven minutes and therefore add further delay.
In comparison to the others low-cost operators, the observed pattern of the first
low cost carrier fits more the profile of a charter carrier.
When looking at a typical hub-and-spoke carrier the differences to low-cost
carriers become apparent:
PDI = 59 %
62
PDI = 48 %
63
5.5
This section presents the results of the analysis of delay propagation on those
sequences for which reactionary delay was reported.
5.5.1 Key factors influencing sequences of reactionary delays
The key factors for the analysis of sequences of reactionary delay need to link
performance in the ground and in the block phase but it is necessary to
differentiate between primary and reactionary delays.
Inbound delay
Root delay
Absorbed delay
Primary delay
Reactionary delay
1
Root
delay
2
Inbound
delay
reactionary
delay
Root delay;
Inbound delay;
Reactionary delay.
64
from 6:00h to 13:59h, afternoon from 14:00h till 21:59h and night from 22:00h till
5:59h.
Figure 34 shows the distribution of the sequences by airline business model.
Nearly 35 percent of all sequences of low-cost operations had a root delay
between one and 15 minutes and occurred in the morning. The main share (
+20%) of these root delays propagated only on one further flight leg (bottom part
of first column of Figure 34).
Summer 2008
1 leg
2 legs
3 legs
>4 legs
40
35
sequences [%]
30
25
20
15
10
5
1-15
16-60
>60
1-15
16-60
>60
1-15
16-60
>60
1-15
16-60
>60
1-15
16-60
>60
1-15
16-60
>60
1-15
16-60
>60
1-15
16-60
>60
1-15
16-60
>60
root delay
Morning Afternoon
[min]
low-cost
Low-cost
Night
Morning Afternoon
Night
Morning Afternoon
Night
Point-to-point
Point-to-point
65
Sequences starting at night time account for about five percent of low-cost
operations, whereas they are barely seen among traditional scheduled flights.
In terms of occurrence, root delays larger than 60 minutes play only a minor role.
They only account for six to eight percent of all sequences.
However there is a difference between the occurrence and the impact on airlines.
Figure 35 illustrates that the impact of the sequences (in terms of reactionary
delay minutes) is distributed quite differently.
Sequences starting in the morning have the biggest impact in terms of reactionary
delay minutes. This corresponds to the high number of sequences in the morning,
which also propagate longer.
On the other hand it is important to notice that long sequences have a big impact,
despite little frequency and/or little root delay.
Summer 2008
1 leg
35
2 legs
3 legs
>4 legs
30
25
20
15
10
5
Afternoon
low-cost
Low-cost
Night
Morning
Night
>60
1-15
Afternoon
16-60
>60
16-60
>60
Morning
1-15
1-15
16-60
>60
16-60
>60
Afternoon
1-15
1-15
16-60
>60
16-60
>60
1-15
1-15
16-60
>60
16-60
>60
Morning
1-15
1-15
root delay
[min]
16-60
Night
Point-to-point
Point-to-point
66
67
delays can end because all root delay is absorbed or because the sequence of the
aircraft ends.
The first chart in Figure 36 shows
minutes,
aircraft
add
around
40
35
30
25
20
15
10
5
0
Dep
delay.
Arr
Dep
Root
absorbed
during
25
20
15
10
5
0
Dep
Arr
Dep
Arr
Dep
Arr
Dep
Arr
Dep
4
90
80
70
60
50
40
30
20
10
0
Dep
Arr
Root
Dep
Dep
Arr
Dep
Arr
Dep
4
160
140
mean departure
delay [min]
120
100
80
60
40
20
0
Dep
Root
Arr
Arr
Dep
1
Arr
Dep
2
Arr
Dep
Arr
Dep
4
Inbound delay
primary delay
absorbed inbound delay, newly added
reactionary delay
Root delay
40
Dep
4
30
Root
reaches
Arr
35
value
Dep
40
the
Arr
45
delay
Dep
more
Arr
1
2
3
sequence of reactionary delay
68
The mean root delay of the third graph equals 81 minutes. The difference to the
previous charts is obvious. Airlines start to absorb the delay right away, with fewer
minutes absorbed during the block phases and more during the ground phases.
They are able to mitigate the root delay despite additional primary delay (yellow
parts).
Sequences with root delays between 121 and 180 minutes show the actual
potential of hub-and-spoke operators to absorb delay. The mean departure delay
can be reduced from 144 to only 61 minutes. It seems that the higher the average
delay, the more are airlines able to avoid further primary delay, and the more they
are able to absorb existing delay.
Looking at all the various charts in Figure 36, is appears that reactionary delays
only increase until a certain level is reached.
Figure 37 evaluates the depth of sequences of hub and spoke carriers with a root
delay between 16 and 60 minutes in summer 2008. As illustrated in Figure 34 and
Figure 35), this group has the highest impact in terms of minutes of reactionary
delay.
When the delay propagates only for one leg, only a small amount is absorbed
during the block and the turn-around phase. The aircraft departs with 25 minutes
of delay, of which 20 minutes are propagated.
When root delay propagates for two legs, the root delay cannot be reduced during
the first block-phase. During the first turn-around phase, about seven minutes are
absorbed but twice as many minutes are added than on sequences with only one
flight leg.
69
During
the
second
turn-around
35
mean departure
delay [min]
40
30
25
20
15
10
5
0
Departure
Arrival
Departure
Root
35
mean departure
delay [min]
40
30
25
20
15
10
5
0
Departure
not
sequences
with
the
a
depth
root
25
20
15
10
5
0
Root
of
delay
Inbound delay
primary delay
absorbed inbound delay, newly added
reactionary delay
Root delay
35
30
Departure
Departure
mean departure
delay [min]
40
Consequently,
Departure
Arrival
is
Departure
sequence
Arrival
2
Arrival
Departure
of
Departure
1
Arrival
depth
Arrival
Root
70
30
25
20
15
10
5
Root
by
long
root
delays
dep
arr
dep
arr
dep
arr
dep
arr
departure
and
50
45
40
35
30
25
20
15
10
5
Root
dep
arr
dep
arr
dep
arr
arr
departure
dep
50
40
30
20
10
Root
dep
arr
dep
dep
arr
140
120
100
80
60
40
20
Root
phase.
arr
arr
3
dep
0
dep
turn-around
160
dep
arr
in
dep
delay
minutes.
absorb
arr
dep
Therefore,
60
departure
scheduling.
70
arr
is
arr
airlines
80
departure
low-cost
of
90
inbound delay
primary delay
absorbed delay, newly added
reactionary delay
root delay
71
departure delay.
The same pattern is seen for all
different
depths
of
40
35
30
25
20
15
10
5
0
reactionary
departure
arrival
departure
Root
sequences.
Low-cost
operators
40
35
30
25
20
15
10
5
0
departure
arrival
Root
departure
arrival
departure
2
mean departure
delay [min]
40
35
30
25
20
15
10
5
0
delay.
departure
arr
Root
78
percent
of
the
analysed
dep
1
arr
dep
arr
dep
3
inbound delay
primary delay
root delay
72
35
30
25
20
15
10
5
Root
Departure
Arrival
Departure
Arrival
Departure
Departure
Arrival
0
Departure
in
Arrival
sequences
Finally,
45
40
35
30
25
20
15
10
5
Departure
Arrival
Departure
Arrival
Departure
2
Arrival
Root
Departure
Arrival
Departure
70
60
50
40
30
20
10
point-to-point
Departure
Arrival
Departure
Arrival
20
0
Departure
Arrival
40
Departure
hub-and-spoke
60
Arrival
delay,
80
Departure
root
100
Arrival
minute
120
Departure
140
Arrival
Departure
160
Arrival
37),
Root
Departure
(Figure
around
Root
Departure
Root
Departure
Arrival
Inbound delay
primary delay
20 minutes less.
Root delay
reactionary delay
73
hub-and-spoke operators. The yellow and orange parts of the columns suggest
that overall point-to-point operators do not add as much primary delay in the turnaround phase as hub-and-spoke operators.
Figure 41 evaluates the depth of
sequences
of
point-to-point
operations
with
root
delay
35
30
25
20
15
10
5
0
Departure
Arrival
Departure
Root
40
35
block phase and during the turnaround phase. After adding another
30
25
20
15
10
5
0
Departure
Arrival
Root
Departure
2
35
Arrival
of delay, on average.
Departure
1
30
25
20
15
10
5
the
absorbed
delay
0
Departure
Arrival
Departure
Root
is
Arrival
Departure
Arrival
Departure
3
Inbound delay
primary delay
Root delay
74
Sequences with three affected legs, depart after the first turn around phase with
an even higher delay than on arrival. During the second and third turn around
phases they absorb another seven and eight minutes of inbound delay and the
aircraft departs on the last flight leg with an average delay of 31 minutes.
The charts reveal that point-to-point operations react quite sensitive to primary
delay. The limited ability to absorb delay, in the turn around and block-to-block
phase, puts more weight on additional delay.
5.5.3.4 First reaction to short departure delays
After an initial root delay, the first opportunity for an airline to react is the following
block phase. In Figure 36 to Figure 41 it was observed that airlines, despite their
overall negative DDI-F, generally add further delay in the block phase following a
rather short departure delay.
Summer 2008
6
4
2
0
-2
-4
-6
1-15 min
16-60 min
>180 min
Hub-and-spoke
Point-to-point
Figure 42: The first reaction after the root delay DDI-F
Figure 42 illustrates the first reaction of airlines irrespective of the number of
sequences in summer 2008 for each of the three airline business models.
It is striking, how the effort to absorb delay in the first block phase increases as the
duration of the initial root delay goes up. Consistent with the observed mean
values of the DDI-F, low-cost airlines are able to absorb more delay than hub-and-
75
spoke or point-to-point airlines. The overall mean values of the three operations
are only reached for root delays of more than 120 minutes.
5.5.3.5 Sequences in Summer 2008 and Winter 2008-09
Figure 20 in section 5.2.2 shows that the ratio of reactionary to primary delay of
hub-and-spoke operations decreased noticeably from winter 2008-09 to summer
2009. Along with the drop of the ratio, the mean delay per delayed departures
also decreased. While primary delays decreased only moderately from eleven to
nearly ten minutes, reactionary delays dropped from eight to below six minutes, on
average.
Figure 40 shows a typical, quite frequent sequence with a mean root delay of 32
minutes and three affected flight legs for hub-and-spoke operations. The
comparison of the winter 2008 season to the summer 2009 season confirms the
45
40
35
30
25
20
15
10
5
0
mean departure
delay [min]
mean departure
delay [min]
Dep
Root
Arr
Dep
Arr
Dep
Arr
1
2
sequence of reactionary delay
Dep
45
40
35
30
25
20
15
10
5
0
Dep
Root
primary delay
Root delay
Arr
Dep
Arr
Dep
Arr
1
2
sequence of reactionary delay
reactionary delay
Dep
3
Inbound delay
76
40
Root
1
2
sequence of reactionary delay
Root
Departure
Departure
Departure
Arrival
Departure
Arrival
15
Departure
15
Arrival
20
Departure
20
Arrival
25
Departure
25
30
Arrival
30
35
Departure
mean departure
delay [min]
35
mean departure
delay [min]
Arrival
40
primary delay
Root delay
reactionary delay
Inbound delay
Figure 44: Sequences with root delays between 16-60 minutes during
the morning and afternoon (Hub-and-spoke operations)
Figure 44 demonstrates that the mean root delay is similar in the morning and in
the afternoon. Although there is a little less primary delay (yellow and orange)
during the turn-around phase in the afternoon, the mean departure delay is higher.
The difference results mainly from fewer absorbed delay minutes, especially in the
turn-around phase.
This indicates that the increasing ratio in the afternoon (see Figure 23) is not only
the result of ongoing delay propagation from root delays in the morning, but also
from a higher level of delay propagation on afternoon-sequences.
77
145
130
mean departure
delay [min]
mean departure
delay [min]
115
100
85
70
55
40
Departure
Root
Arrival
Departure
Arrival
Departure
145
130
115
100
85
70
55
40
Departure
1
2
sequence of reactionary delay
Root
primary delay
Root delay
Arrival
Departure
Arrival
Departure
1
2
sequence of reactionary delay
reactionary delay
Inbound delay
Figure 45: Sequences with root delays between 121-180 minutes during
the morning and afternoon (Hub-and-spoke operations)
Figure 45 shows sequences with a root delay between 121 and 180 minutes and
two affected legs. Along the morning-sequence the aircraft absorbs more than 45
minutes in each turn-around phase (difference between inbound delay and
reactionary delay (green part) but adds 12 and 9 minutes of new primary delay
respectively. Finally the aircraft departs with 68 minutes delay.
In contrast to the propagation of delay in the morning, the aircraft absorbs on
average less than 20 minutes during the first turn-around phase in the afternoon.
With even less primary delay each in each turn-around phase and 45 minutes of
absorbed delay in the second turn-around phase, the aircraft finally leaves with still
more than 90 minutes of delay.
It becomes evident, that airlines do not absorb as much inbound delay in the
afternoon. The magnitude of the root delay is 40 percent higher in the afternoon
than in the morning (154/144=1,07 in the morning and 210/143=1,47 in the
afternoon).
The same operational difference was observed for point-to-point and low-cost
operations.
5.6
This section analyses the magnitude and the depth of sequences of reactionary
delay. As explained in section 4.3.4, the magnitude of the root delay is a simple
and useful indicator, but it is quite sensitive to the length of the root delay and the
depth of the sequence.
78
8,00
4,0
7,00
6,00
Magnitude
5,00
4,00
2,0
3,00
2,00
1,0
1,00
0,00
0,0
1-15
1-15
16-60
1-15
16-60
Low-cost magnitude
low-cost depth
Hub-and-spoke magnitude
hub-and-spoke depth
Point-to-point magnitude
point-to-point depth
79
3,0
In Figure 35 it was already shown that the impact of all reactionary delay minutes
is lower in the afternoon due to the lower number of rotational frequencies. Figure
46 confirms that sequences are on average shorter in the afternoon, and therefore
have a lower impact in terms of magnitude.
However it should be noted that Figure 40 demonstrated that despite the lower
overall impact, the level of delay propagation is higher in the afternoon.
Due to the small number of flights, the magnitude for flights during night time is
artificially high and should be viewed with a note of caution.
It is interesting to note that the morning-sequences with a root delay between 61120 minutes show the highest mean depth and also the biggest difference in
comparison to the afternoon.
Hub-and-spoke operators show the lowest depth and magnitude. In terms of
ranking between the three business models, the magnitude reflects the ratio of
reactionary to primary delay, analysed in 5.2.2.
Low-cost operators show a surprisingly high magnitude for short root delays,
especially in the morning (5.8). This supports the previous observation that lowcost carriers do only have limited scope to absorb delay and are in fact more likely
to add new primary delay in the turn-around phase. In Figure 42 it was already
illustrated that low-cost operators even add delay during the first block phase of
such a short root delay.
In 5.5.3.1 it was suggested that, the higher the actual level of delay, the more
delay can be absorbed and the fewer newly added primary delay. This is also
reflected in the morning-magnitude in Figure 46. The magnitude drops although
root delay and depth increase. This confirms that the delay propagation is
significantly lower on flights with relatively long root delays.
It is apparent that the magnitude works well as an indicator, when including the
depth of a sequence.
80
5.7
After the analysis of delay propagation by airline type, this section focuses on the
delay propagation at European airports.
5.7.1 Reactionary to primary delay ratio at selected airports
Figure 47 shows the reactionary to primary delay ratio for six major European
airports. All other airports are grouped within others. It should be noted that the
analysis is still based on the validated data sample used for the analysis in the
previous chapter of the report. Due to airline data confidentiality reasons, airports
are dis-identified, as most of the airports have only one major carrier serving the
airport.
Summer 2008
primary delay of delayed
departures
30
25
25
20
20
15
15
10
36%
50%
42%
44%
29%
45%
42%
44%
39%
34%
33%
25%
28%
45%
44%
42%
0
Summer 2009
50%
43%
Winter 2008-09
Summer 2008
10
Winter 2007-08
0
hub other hub other hub other
ops
ops
ops
% of all
departures
AP1
100
80
60
40
20
0
AP2
AP3
hub other
ops
AP4
AP6
other AP
on-time departures
45
19
43
55
52
51
19
14
17
14
38
28
45
delayed departures
81
82
It is striking that one of the biggest hubs in Europe London Heathrow (EGLL)
only affects 29 airports on average. However, it should be noted that the analysis
was restricted to airports within the ECAC area.
Mean number of
daily destinations
Summer 2008
100
90
80
70
60
50
40
30
20
10
0
47
41
39
38
34
29
23
19
10
the
minimum
of
either
the
Root 1. affected
airport
AP
3. affected
AP
Primary delay
Absorbed delay, replaced by new
Rest of reactionary delay
Reactionary delay from root
Inbound delay
Root delay
2. affected
AP
83
Departure
Arrival
0
Departure
20
Arrival
30
Departure
40
Arrival
delay
Departure
reactionary
departure delay
[min]
reported
50
phase is disregarded.
For example, the second airport in Figure 49 suffered an inbound arrival delay of
43 minutes but the total departure delay on the subsequent outbound flight leg
was 41 minutes of which only 35 minutes were carried over from the previous flight
leg. However, only 30 minutes were caused at the root airport. Therefore the
propagated delay due to the initial root delay is the minimum of 43, 35 and 30 in
this case 30 minutes.
In Figure 50 the average daily impact of selected European Hubs - on themselves
and on other airports - is shown. Secondary and other hubs not included in the list
of hub airports are grouped together as secondary airports. All other ECAC
airports were grouped as Other.
84
85
30,0
25,0
20,0
15,0
10,0
5,0
50%
54%
56% 51%
43%
33%
37%
30%
20% 23%
EDDF
EDDM
EGKK
EGLL
EHAM
EKCH
LEBL
LEMD
LIRF
Summer 2008
Winter 2007-08
Summer 2008
Winter 2007-08
Summer 2008
Winter 2007-08
Summer 2008
Winter 2007-08
Summer 2008
Winter 2007-08
Summer 2008
Winter 2007-08
Summer 2008
Winter 2007-08
Summer 2008
Winter 2007-08
Summer 2008
Winter 2007-08
Summer 2008
Winter 2007-08
0,0
LSZH
The share of delay returned to the origin airport varies between 20 percent for
Zurich-Kloten (LSZH) to 56 percent for London-Gatwick (EGKK). In absolute
terms, the average minutes of delay returned to the origin airport range from three
minutes at Zurich Kloten (LSZH) to 12 minutes at Rome-Fiumicino (LIRF) airport.
For Frankfurt (EDDF), London Heathrow (EGLL) and Copenhagen (EKCH) a
notable difference between summer and winter season can be observed.
86
Figure 54 shows the propagation of root delays between 16 and 60 minutes along
the rotational sequences on the 8th of December 2008.
Sequences on which the delay propagated on four successive flight legs, the
observed root delay was 39 minutes in Frankfurt. When the aircraft returned to
87
Frankfurt, the inbound arrival delay was similar to the observed delay when the
flight departed from Frankfurt. Although the aircraft was able to absorb nearly 19
minutes during the turn-around phase, it suffered another long primary delay of
about 64 minutes (yellow and orange part). On the third leg the aircraft was able to
absorb around four minutes, coming back with then 79 minutes of inbound arrival
delay. However, after adding another 33 minutes in Frankfurt, the total departure
delay increased to 98 minutes.
DEPARTURE root delays 16-61 minutes at EDDF
Hub-and-spoke operations 08.12.2008
120
100
80
60
40
20
0
FRA
arr
dep
Root
arr
dep
arr
FRA
dep
arr
dep
FRA
140
120
100
80
60
40
20
0
FRA
Root
arr
dep
1
arr
dep
FRA
arr
dep
arr
dep
arr
FRA
dep
5
arr
dep
FRA
reactionary delay
primary delay
inbound delay
The bottom chart of Figure 54 shows a similar sequence with two additional flight
legs. The sequence started with a lower average root delay of 21 minutes at
Frankfurt. Every time the aircraft returned to Frankfurt, the impact of each turnaround phase became more evident. Each time at Frankfurt the additional
88
89
CONCLUSION
Throughout this analysis various KPIs were introduced to observe and measure
delay propagation. Indicators aimed at measuring airline performance during the
block-to-block and turn-around phase illustrated differences in airline strategies
and formed the basis for the more detailed analysis of the delay propagation along
the individual flight legs.
The ratio of reactionary to primary delays measures the sensitivity to reactionary
delays. For the sample of selected airlines its mean value is slightly below one.
Thus, almost half of the departure delay is due to reactionary delays.
The comparison between calculated and reported reactionary delays revealed that
calculated reactionary delays appear higher than the reported ones because they
do not take additional primary delay during the ground phase into account.
Over the observed four seasons, on average 50 percent (12 minutes) of delays in
low-cost operations are reactionary delays. Hub-and-spoke operators have by far
the lowest ratio as reactionary delays account for early 40 percent of all delays (7
minutes). Point-to-point operations lie in between the other two with around 45
percent of reactionary delay (9 minutes).
KPIs evaluating the turn-around and the block-to-block performance demonstrated
the following:
The BTO shows a strong and linear correlation to the DDI-F. The larger the share
of aircraft which exceed the scheduled block-to-block time, the less delay can be
absorbed in the block-to-block phase. On average, irrespective of the business
model, the DDI-F is negative. Therefore, buffer time is included in the scheduled
block-to-block phase of all types of operation. However, with an average DDI-F of
about minus five minutes, low-cost operators are best positioned to absorb delays
in the block-to-block phase.
Hub-and-spoke operators showed an average DDI-F of three minutes and pointto-point operators a DDI-F of two minutes.
90
The correlation of the GTO to the DDI-G looks similar to that of the BTO and DDIF. Depending on the airline business model, between 60 and 90 percent of all
analysed flights exceed the scheduled turn-around time. However, only half as
many flights exceed their scheduled turn-around times when additional minutes
due the aircraft arriving ahead of its scheduled arrival time are removed.
Finally, the average absorbed inbound arrival delay provided an understanding of
the level of delay that can be absorbed during the turn-around phase. Here, lowcost airlines appeared to have only a limited ability to absorb delay in the turnaround phase. Instead, they even added the highest level of new primary delays.
Overall,
hub-and-spoke
and
point-to-point
carriers
are
able
to
absorb
approximately the same amount of delay during the turn-around phase, but huband-spoke carriers added more new primary delays than point-to-point carriers.
Thus, the ratio of reactionary to primary delay is lower for hub-and-spoke carriers.
Irrespective of the airline business model, the time of the day and the length of the
delay, the majority of the root delays can be recovered within the first leg after the
root delay occurred. Those sequences (with one affected leg) accounted for 50 to
60 percent of all the analysed sequences.
While of the share of sequences with a root delay between one and 15 minutes
accounts for the majority of observed sequences, the impact in terms of
reactionary delay minutes is the highest for root delays between 16 and 60
minutes. As can be expected, sequences starting in the morning have the most
sever impact on reactionary delays and account for about 60-65 percent of all
reactionary delay.
Depending on the airline business strategy notable differences in strategies to
mitigate reactionary delay were observed.
Hub-and-spoke operations show a limited reduction of reactionary delay for short
root delays between 1 and 15 minutes. In fact, sequences with a short root delay
are likely to add new primary delay on subsequent flight legs which further
increases the overall level of reactionary delay. The reaction on longer root delays
(>120 min.) is very different. Aircraft are able to absorb a significant amount of
delay in each turn-around phase and manage to avoid additional primary delays
91
92
93
OUTLOOK
In this study, reporting issues and uncertainties represented the greatest challenge
while dealing with the data. EUROCONTROL and IATA are working on an
appropriate, adjusted framework for reporting delays.
A set of more specific but comprehensive delay codes needs to be developed in
order to separate delay causes more clearly from another. Many major airlines
already use subcategories within their internal delay code scheme. A general
guideline and/or instructions applicable to all airlines need to be developed.
Additionally, a very simple local quality check at the Operations Control Centre
would help to further improve the quality of the data. An automatic warning should
be generated if sum of individual delays reported for a flight exceeds the total
departure delay or when rotational reactionary delays is larger than the reported
inbound arrival delay.
All this would ensure the validity of results, reducing a possible bias from airline
coding policies.
For the analysis of delay propagation, the reporting of callsigns which cause nonrotational reactionary delays is of upmost importance. If airlines started to report
these callsigns, a whole new analysis addressing the actual network effect of
delay propagation could be worked out.
These callsigns would also enable the analysis of relations and impacts of delay
propagation within airline alliances regarding the magnitude of delay propagation
and consequently the costs caused by the respective alliance partners.
Whether the propagation of long delays is preferred over cancelling flights is
unknown at this point and factors influencing this decision probably vary from
airline to airline. Obviously, this decision has an overall impact on the propagation
of delays. Therefore, different cancellation strategies should be looked at and
compared. If original aircraft rotations were provided by an airline it could be
compared to the actual operated rotation. Then the impact of swapping and/or
cancelling flights within the fleet of an airline as well as the entire network can be
analysed.
94
As a follow on to this study, various IF-cases could be tracked and analysed with
the created sequences of reactionary delays.
For example,
95
GLOSSARY
ADDD
Afternoon
ANS
ANSP
ATFM
ATFM delay
(CFMU)
ATFM Regulation
ATM
Bad weather
Block time
CDM
CET
CFMU
CODA
CRCO
DST
EATM
ECAC
E-CODA
EET
ETFMS
EU
EUROCONTROL
EUROCONTROL
Member States
96
GMT
Ground phase
IATA
ICAO
IFR
KPI
Morning
MVT
Night
OCC
OOOI-times
PDD
PRC
Primary Delay
PRISME
PRU
Punctuality
Reactionary delay
Root delay
Slot (ATFM)
SGT
STA
STATFOR
STD
UTC
97
BIBLIOGRAPHY
Ahmad Beygi, S., Cohn, A., Guan, Y. and Belobaba, P. (2008): Analysis of the
potential for delay propagation in passenger airline networks. In Journal of Air
Transport Management, Vol. 14, Pp. 221-236.
Bazargan, M. (2004): Airline operations and Scheduling. Burlington, USA, Ashgate
publishing company.
Beatty, R., Hsu, R., Berry, L. and Rome, J. (1998): Preliminary evaluation of flight
delay propagation through an airline schedule. In 2nd USA/Europe air traffic
management r&d seminar, Orlando, 1.-4.12.1998.
CODA (2009): Delays to Air Transport in Europe. DIGEST Annual 2008, Brussels,
Belgium, EUROCONTROL.
CODA homepage (EUROCONTROL): https:\\extranet.eurocontrol.int\http:\\prismeweb.hq.corp.eurocontrol.int\ecoda\portal. 23.October 2009.
Cook, A. (2007): European Air Traffic Management - Principles, Practice and
Research. Burlington, USA, Ashgate publishing company.
CRCO homepage (EUROCONTROL):
http://www.eurocontrol.int/crco/public/subsite_homepage/homepage.html.
23.October 2009.
Diana, T. (2009): Do market-concentrated airports propagate more delays than
less concentrated ones? A case study of selected U.S. airports. In Journal of Air
Transport Management, Vol. 15, pp.280-286.
ECAC webpage: http://www.ecacceac.org/index.php?content=lstsmember\&idMenu=1\&idSMenu=10. 11.November
2009.
EUROCONTROL Experimental Centre (2003): Flight Delay Propagation.
Synthesis of the Study. EEC Note No 18/03, Brussels, Belgium, EUROCONTROL
Gillen, D., Hansen, M. M. and Djafarian-Tehrani, R. (2000): Aviation Infrastructure
Performance and Airline Cost: A statistical Cost Estimation Approach. Wilfird
Laurier University and Institute of Transportation Studies, Institute of
Transportation Studies, National Center of Excellence in Aviation Operations
Research, University of California at Berkeley, Berkeley, USA, Elsevier Science
Ltd.
Guest, T. (2007): Air traffic delay in Europe. Trends in Air Traffic Vol. 2, Brussels,
Belgium, EUROCONTROL.
IATA (2001): IATA Airport Handling Manual. 21st Edition, Montreal, Canada.
98
99
100
101
102
Cy
CallSign
IACO 3-letter flightnumber prefix followed by the flight number (no blanks)
ComFltNbr
AcReg
Dep ICAO
4-letter code of the departure station (the IATA 3-letter code can also be
accepted)
Dst ICAO
4-letter code of the destination station (the IATA 3-letter code can also be
accepted)
Std
Sta
Eet (FP)
Out
Off
On
In
Dl1
Time1
Dly2
Time2
Dly3
Time3
Dly4
Time4
Dly5
Time5
RD
from Flt If there is a reactionary delay, give the call sign of the flight
having directly caused the reactionary delay
STXO
STXI
ServType
FltType
QC
Quality Control ("A" for ACARS, "M" for Manual or "C" for
Combination or both)
103
Winter
Summer
GMT = UTC
GMT = UTC + 1h
CET = UTC + 1h
CET = UTC + 2h
EET = UTC + 2h
EET = UTC + 3h
GMT
country
Ireland
United Kingdom
Portugal
Canary Islands,
Spain
Faroe Islands,
Denmark
CET
ICAOCode
EI
EG
LP
country
Albania
Austria
Belgium
BosniaGE, GC Herzegovina
EKFO
Croatia
Czech Republic
Denmark
France
Germany
Hungary
Italy
Kosovo,
Montenegro,
Serbia
Luxemburg
Macedonia
Malta
Monaco
Netherlands
Norway
Poland
Slovakia
Slovenia
Spain
Sweden
Switzerland
104
EET
ICAOCode
LA
LO
EB
country
Bulgaria
Cyprus
Estonia
ICAOCode
LB
LC
EE
LQ
Finland
EF
LD
LK
EK
LF
ED
LH
LI
Greece
Latvia
Lithuania
Moldova
Romania
Turkey
LG
EV
EY
LU
LR
LT
LY
EL
LW
LM
LN
EH
EN
EP
LZ
LJ
LE
ES
LS
105
B190
JS32
E120
D328
SF34
E135
AT43
AT45
CRJ1
CRJ2
DH8C
E145
F50
SB20
AN26
AT72
CRJ7
E170
DH8D
RJ70
F70
RJ85
CRJ9
B462
RJ1H
F100
E190
B463
MD87
B735
B736
A318
A319
MD82
B733
B737
MD83
MD88
B734
MD81
A320
MD90
B738
B739
B752
A321
B762
A310
B763
A30B
A332
A343
A306
B772
MD11
A333
A346
B773
B744
B742
19
19
30
32
34
37
46
46
50
50
50
50
50
50
52
66
70
70
72
79
80
82
86
88
97
101.5
108
110
110
111
112
114
124
131
137
137
140
142
144
147
150
150
167
178
183
186
198
207
229
240
251
264
266.5
283
285
298
339
380
390
398
42
35
38
40
48
39
39
44
41
38
40
42
33
46
85
43
43
50
44
41
47
41
45
45
44
51
52
50
48
50
47
60
55
53
50
54
48
56
54
44
57
.
60
69
64
62
122
89
98
117
113
116
63
124
120
129
117
134
130
.
106
DECLARATION
I, Martina Jetzki, declare that I have developed and written the enclosed thesis
entitled The propagation of air transport delays in Europe entirely by myself and
have not used sources or means without declaration in the text. Any thoughts or
quotations which were inferred from these sources are clearly marked as such.
This thesis was not submitted in the same or in a substantially similar version, not
even partially, to any other authority to achieve an academic grading and was not
published elsewhere.
Brussels, 23.12.2009
107